TR Forums

Hey, guys. I've been using my FX-8350 for a while now and as always, I'm quite curious as to how the cores are numbered. You know, the geek in me is just dying to know where exactly on the Vishera die a particular Core # is. There's this marketing slide AMD so graciously gave us but the guy who did it obviously got a little lazy or didn't understand a thing the engineers told him (yeah, 'cuz he's from marketing).

So, any better info out there? Last time I emailed and asked AMD about core # in my old Phenom II, he searched the Internet for a photo. Funny, aren't they supposed to be the source for all these AMD-related infos? Just didn't expect it.

The problem here is, even the dieshot tells you those core "numbers", that is not necessary how the BIOS/OS (and I am not sure if say Windows or Linux uses the same "number assignments") will see them. It may be totally reversed, or just random. The only surefire way to find out is probably with a higher precision infrared scanner, and then you run an app that use core affinity and exercise each core to 100% and look at the heat signatures.

It had better not be completely random. The thread scheduler in the OS needs to be aware of which cores share resources (due to the Bulldozer/Piledriver "module" architecture), in order to make optimal scheduling decisions. But beyond that, yeah there's nothing that says things need to be numbered sequentially. There might even be some advantage to making it non-contiguous at the module level, so that if you're (say) loading cores 1-4 the active cores are at opposite ends of the die, thereby spreading out heat production and reducing the chance of hot spots.

@Flying Fox:

I was afraid you'd say that. I have a feeling those pretty die shots are nothing more than marketing material and the core numbering is there just because. When I asked AMD for core numbering info about Phenom II, the tech support guy just pulled a picture from somewhere in the Net. I can't remember clearly but it seemed like there was something fishy about his response.

I actually once thought about what you said, about just getting some sort of infrared heat sensor and track which part of the die is heating up while loading cores individually, but those things are pricey. Besides, how would you do it with the heatsink on?

@JBI:

Yes, I'd imagine so. I once considered core numbering to 'reset' everytime the system turned on or restarted, but that would be a bit messy, wouldn't it? I also imagine the BIOS needs to start off with a 'default' core first, probably Core 0, assign the OS to this 'default' core during boot, then things could just spread out to other cores from there.

ronch wrote:

Besides, how would you do it with the heatsink on?

Run the mobo in a tank of mineral oil with no HSF.

With the heatsink off, put several small vessels filled with ethanol on top of the CPU, and see where ethanol evaporates first. This is how you spot small differences in temperature.

ronch wrote:

I was afraid you'd say that. I have a feeling those pretty die shots are nothing more than marketing material and the core numbering is there just because. When I asked AMD for core numbering info about Phenom II, the tech support guy just pulled a picture from somewhere in the Net. I can't remember clearly but it seemed like there was something fishy about his response.

IMO it is not that fishy, it is implementation details that are just not to be disclosed. High resolution die shots already give competitors enough insight into the implementation (and large enough competitors will use scanning electron microscope to augment their reverse engineering efforts).

ronch wrote:

I actually once thought about what you said, about just getting some sort of infrared heat sensor and track which part of the die is heating up while loading cores individually, but those things are pricey. Besides, how would you do it with the heatsink on?

I thought you just need to put the sensor in the right distance and it will probe properly? I never use those things so I can't say for sure.

ronch wrote:

Yes, I'd imagine so. I once considered core numbering to 'reset' everytime the system turned on or restarted, but that would be a bit messy, wouldn't it? I also imagine the BIOS needs to start off with a 'default' core first, probably Core 0, assign the OS to this 'default' core during boot, then things could just spread out to other cores from there.

Probably a combination of internal implementation, microcode on the CPU, and BIOS code.

Ok, I emailed AMD and asked them for it. Here's their response:

Thank you for contacting AMD!

I understand you are looking for some technical documentation on the FX-8350.

Please be advised that AMD holds the right to not provide proprietary technical documentation to the general public. Unless you are a major embedded partner of AMD, chances are you will not gain any access to this information.

Please go to this link for specific technical documentation to the general public.

http://support.amd.com/us/Pages/techdocs.aspx

You can try reviewer sites like Tomshardware.com, Anandtech.com, and Hardwarecanucks.com to find out some of the diagrams that show the CPU die.

Chances are I will not gain any access to this information. You gotta love that part.

If I had the chance to rate this response I'd probably give it 0/10.